I'm Wes — a Senior Software Engineer with a strong foundation in Computer Science and IT (NJIT), complemented by an advanced Full-Stack Web Development certification from Rutgers. With 7+ years of industry experience, I've spent the last 3 years specializing in building and fine-tuning LLM-based platforms using C++, Python, TypeScript, React, Next.js, Pinecone, and Postgres to deliver scalable, knowledge-driven systems.
What I've Done...
Due to privacy and contractual obligations, I’m unable to share the actual client-facing platforms I’ve built. Instead, this portfolio features representative projects that reflect core functionalities of the real-world systems I’ve developed—demonstrating how I translate complex requirements into scalable, effective solutions.
Challenges & Solutions
Here’s how my teammates and I addressed common issues faced while building advanced, scalable platforms.
Challenge: Retrieval-augmented AI agents were generating inaccurate or hallucinated responses due to lack of grounding in verified business knowledge bases.
Solution: Implemented a Python-based RAG system with domain-specific indexing and vector search, reducing hallucination rates by over 80% (from ~18-25% industry baseline to 2-3%).
Challenge: During LLM fine-tuning, many challenge sets were failing due to ambiguous prompts or insufficient coverage across edge cases.
Solution: Redesigned challenge structure with clear intent-labeling, comprehensive test cases using PyTest, and JSON-based input/output validation—significantly increasing model success rates and review scores.
Challenge: The team lacked a unified frontend system, resulting in inconsistent UI patterns and duplicate logic across React components.
Solution: Developed a shared component library in Storybook with integrated Redux Toolkit logic, streamlining design consistency and reducing dev time by 40%.
Challenge: New internal tools were being blocked by rigid monoliths and legacy API endpoints that weren’t suited for multi-team dev workflows.
Solution: Refactored the architecture into micro-frontends using Module Federation and RabbitMQ for async task processing—enabling scalable, decoupled deployment across multiple departments.
Challenge: Retrieval-augmented AI agents were generating inaccurate or hallucinated responses due to lack of grounding in verified business knowledge bases.
Solution: Implemented a Python-based RAG system with domain-specific indexing and vector search, reducing hallucination rates by over 80% (from ~18-25% industry baseline to 2-3%).
Challenge: During LLM fine-tuning, many challenge sets were failing due to ambiguous prompts or insufficient coverage across edge cases.
Solution: Redesigned challenge structure with clear intent-labeling, comprehensive test cases using PyTest, and JSON-based input/output validation—significantly increasing model success rates and review scores.
Challenge: The team lacked a unified frontend system, resulting in inconsistent UI patterns and duplicate logic across React components.
Solution: Developed a shared component library in Storybook with integrated Redux Toolkit logic, streamlining design consistency and reducing dev time by 40%.
Challenge: New internal tools were being blocked by rigid monoliths and legacy API endpoints that weren’t suited for multi-team dev workflows.
Solution: Refactored the architecture into micro-frontends using Module Federation and RabbitMQ for async task processing—enabling scalable, decoupled deployment across multiple departments.
About Me
What inspires me...
Ikigai: The Japanese Secret to a Long and Happy Life by Héctor García & Francesc Miralles
Explore the technologies and tool I use to craft digital solutions...
Explore my interests and hobbies beyond the code...
Contact Me
How I can help...